Mobile edge computing has emerged as a promising paradigm to complement the computing and energy resources of mobile devices. In this computing paradigm, mobile devices offload their computing tasks to nearby edge servers, which can potentially reduce their energy consumption and task completion delay. In exchange for processing the computing tasks, edge servers expect to receive a payment that covers their operating costs and allows them to make a profit. Unfortunately, existing works either ignore the payments to the edge servers, or ignore the task processing delay and energy consumption of the mobile devices. To bridge this gap, we propose an auction to allocate edge servers to mobile devices that is executed by a pair of deep neural networks. Our proposed auction maximizes the profit of the edge servers, and satisfies the task processing delay and energy consumption constraints of the mobile devices. The proposed deep neural networks also guarantee that the mobile devices are unable to unfairly affect the results of the auctions. Our extensive simulations show that our proposed auction mechanism increases the profit of the edge servers by at least 50% compared to randomized auctions, and satisfies the task processing delay and energy consumption constraints of mobile devices.

移动边缘计算已成为一种有前途的范例，可以补充移动设备的计算和能源资源。在这种计算范例中，移动设备将其计算任务转移到附近的边缘服务器，这有可能减少其能耗和任务完成延迟。作为处理计算任务的交换，边缘服务器希望获得一笔支付，以支付其运营成本并使其获利。不幸的是，现有工作要么忽略了向边缘服务器的付款，要么忽略了任务处理的延迟和移动设备的能耗。为了弥合这一差距，我们提出了拍卖活动，以将边缘服务器分配给由一对深度神经网络执行的移动设备。我们建议的竞价将边缘服务器的利润最大化，并满足了移动设备的任务处理延迟和能耗约束。提出的深度神经网络还保证了移动设备不会不公平地影响拍卖结果。我们广泛的模拟表明，与随机拍卖相比，我们提出的拍卖机制将边缘服务器的利润提高了至少50％，并且满足了移动设备的任务处理延迟和能耗限制。